Pedotransfer functions (PTFs) are widely used in science and engineering to calculate soil thermal diffusivity () as a function of soil moisture content and other soil physical properties. However, available data on are rather scarce and already developed PTFs suffer from significant uncertainties in applicability and performance. Few studies attempted to systematically review and evaluate these PTFs with large dataset of . The objectives of this study were to (1) review and collate currently available PTFs for ; (2) compile a dataset to evaluate performance of these PTFs, (3) build a machine learning (ML) (i.e., random forest-RF) model with three classes of soil physical properties to determine soil physical properties most appropriate to use as predictors for calculating and (4) compare the RF modelling with the traditional PTFs. The correlation coefficient (), centered root-mean-square (), and standard deviation (SD) were used to evaluate the performance of the above PTFs. A total of nine PTFs were synthesized and assessed with a compiled dataset consisting of 998 measurements from 106 soils with a wide range of textures. In general, the PTFs of Lukiashchenko and Arkhangelskaya (2018) (LA2018–1 and LA2018–2) perform the best. While the quadratic equation of Mady and Shein (2018) (MS2018–1 and MS2018–2) showed better performance for coarse and medium textured soil. However, the overall performance of the nine PTFs for predicting was not as accurate as RF. RF can satisfactorily calculate taking into account sand, silt, clay contents, and soil moisture content as predictors.

中文翻译：

---

估计土壤热扩散率的随机森林方法：与传统土壤传递函数的评估和比较


Pedotransfer 函数 (PTF) 广泛应用于科学和工程中，用于计算土壤热扩散率 () 作为土壤水分含量和其他土壤物理特性的函数。然而，可用的数据相当稀少，并且已经开发的 PTF 在适用性和性能方面存在很大的不确定性。很少有研究尝试使用大型数据集系统地审查和评估这些 PTF。本研究的目标是 (1) 审查和整理当前可用的 PTF； (2) 编译数据集以评估这些 PTF 的性能，(3) 构建具有三类土壤物理特性的机器学习 (ML)（即随机森林-RF）模型，以确定最适合用作预测变量的土壤物理特性用于计算并 (4) 将 RF 建模与传统 PTF 进行比较。使用相关系数（）、中心均方根（）和标准差（SD）来评估上述PTF的性能。总共合成了 9 个 PTF，并使用编译的数据集进行了评估，该数据集由 106 种不同质地土壤的 998 个测量值组成。总的来说，Lukiashchenko 和 Arkhangelskaya (2018)（LA2018-1 和 LA2018-2）的 PTF 表现最好。而 Mady 和 Shein (2018) 的二次方程（MS2018-1 和 MS2018-2）对于粗质和中质土壤表现出更好的性能。然而，九个 PTF 的整体预测性能不如 RF 准确。 RF 可以将沙子、淤泥、粘土含量和土壤水分含量作为预测因子进行令人满意的计算。